Workify
A minimal tool for creating web workers APIs, weighing 740 bytes (430b gzipped).
npm install @nnilky/workify
https://www.npmjs.com/package/@nnilky/workify
Example
import { createMessageHandler, type InferInterface } from "@nnilky/workify";
const add = (a: number, b: number) => a + b;
const handler = createMessageHandler({ add });
export type Interface = InferInterface<typeof handler>;
onmessage = handler;
import { createWorker } from "@nnilky/workify";
import type { Interface } from "./worker";
const workerUrl = new URL("./worker.ts", import.meta.url)
const [api] = createWorker<Interface>(workerUrl);
const result = await api.add(1, 2);
console.log(`1 + 2 = ${result}`);
Getting the Worker URL
Getting the worker to bundle correctly is a bit finicky.
If you are using Vite, I recommend using their url import syntax
import workerURL from "./worker?worker&url";
const [api] = createWorker(workerURL);
Otherwise, the recommended way is to use new URL("./path-to-worker", import.meta.url)
const workerURL = new URL("./worker.ts", import.meta.url);
const [api] = createWorker(workerURL);
However, please consult your bundlers documentation for proper instructions:
Worker Pool
You can construct a worker pool the same way you'd make a worker. You can optionally specify the number of workers to use with the default being navigator.hardwareConcurrency
.
import { createWorkerPool } from "@nnilky/workify";
import type { Interface } from "./worker";
const [api] = createWorkerPool<Interface>(new URL("./worker.ts", import.meta.url));
const promises = []
for (let i = 0; i < 16; i++) {
promises.push(api.renderFrame(i))
}
const frames = await Promise.all(promises)
This just redirects each function call to a different worker round robin style.
Transfers
In order to transfer objects to and from workers, use transfer()
. You can only transfer types that are Transferable.
import { transfer } from "@nnilky/workify";
const [api] = createWorker(new URL("./worker.ts", import.meta.url));
const canvas = new OffscreenCanvas(100,100)
const image = canvas.transferToImageBitmap()
transfer(image)
api.resizeImage(image)
import { transfer } from "@nnilky/workify";
const createImage = () => {
const canvas = new OffscreenCanvas(100,100)
const image = canvas.transferToImageBitmap()
transfer(image)
return image
}
const handler = createMessageHandler({ createImage });
export type Interface = InferInterface<typeof handler>;
onmessage = handler;
This works under the hood by creating a list of values that are included in the transfers in the next request/reponse.
Because of this, It's critical you do this right before sending a request/returning a response. This to avoid any race conditions caused by sending those objects with different request/response.
const image = await createImage()
transfer(image)
const thumbnail = await generateThumbnail(image)
transfer(thumbnail)
return { image, thumbnail }
const image = await createImage()
const thumbnail = await generateThumbnail(image)
transfer(image)
transfer(thumbnail)
return { image, thumbnail }
Cleanup
In order to terminate workers when you don't need them, createWorker
and createWorkerPool
both return the actual workers as their second return value. You can use this in your framework to perform cleanup.
For example, in Solid.js it would look something like this:
import { type WorkerInterface, createWorker, createWorkerPool } from "@nnilky/workify"
const useWorker = <T extends WorkerInterface>(url: string): T => {
const [api, worker] = createWorker<T>(url);
onCleanup(() => worker.terminate())
return api
}
const useWorkerPool = <T extends WorkerInterface>(url: string): T => {
const [api, workers] = createWorkerPool<T>(url);
onCleanup(() => workers.forEach(v => v.terminate()))
return api
}
Customising Worker
If you want to have a non-request/response protocol on your worker, for instance reporting back updates to a task. You can add an if condition to the onmessage
global for change how data is recieved.
import { createMessageHandler, type InferInterface } from "@nnilky/workify";
import { performLongTask } from "./custom";
const add = (a: number, b: number) => a + b;
const handler = createMessageHandler({ add });
export type Interface = InferInterface<typeof handler>;
onmessage = (e) => {
if (e.data.isCustomRequest) {
runCustomRequest(e)
} else {
handler(e);
}
}
Or if you want to send data back, such as a progress report, you can just send it using a normal postMesssage
call and attach an message listener to the exposed worker, workify will ignore any non-workify messages.
postMessage({ type:"progress-update", progress: 0 })
const [api, worker] = createWorker(workerURL)
worker.addEventListener("message", (e) => {
if (e.data?.type !== "progress-update") return
console.log(`Progress: ${e.data.progress}`)
})
How it works
Under the hood, when you try to call a method on a worker, the reference to the function is proxied. Only the function name and arguments are sent to the worker, this is then recieved on the other end and mapped to the correct function.
The Interface
generic lets you have a usable developer experience by providing proper typing to the proxy object.